01
Define
Declare dependencies, retry policy, and outputs in readable Lua.
- A webhook and CLI command start a flow.
- User and order requests run independently.
- Results join at a routing step.
- High-value, standard, and bulk branches run in parallel.
- A failed task retries before the flow stores its result.
One language. One runtime. One visible path.
Define, run, and observe workflows
02
Run
Execute the same flow from a shell, an API request, or a named webhook.
03
Observe
Follow run and task lifecycle events without surrendering your infrastructure.
local flow = Flow.new("nightly_report")
flow:step("users", nodes.http_get({ url = users_url }))
flow:step("orders", nodes.http_get({ url = orders_url }))
flow:step("report", function(ctx)
return build_report(ctx.users_data, ctx.orders_data)
end):depends_on("users", "orders")The runtime does the hard part
Branch.
Retry.
Rejoin.
- 00:00.000Fetch usersComplete
- 00:00.000Fetch ordersComplete
- 00:00.184Generate reportRetry 2/3
- 00:02.301Store resultComplete
Independent work runs in parallel. Dependencies wait. Retry and timeout policy stays with the flow—not scattered through every task.
Built-in capability
103 nodes.
Zero drag.
HTTPTransformDocumentS3AIDatabaseMCPShell
When the flow needs judgment
Hand it to a crew.
Some steps can't be encoded: research, interpretation, a decision someone must stand behind. IronCrew — IronFlow's sibling runtime, the same single-binary Rust + Lua shape — runs specialist agent crews with human questions and fail-closed approval gates. The flow hands over one step, gets a structured decision back over plain HTTP, and keeps moving.
Start where you are
One command.
The flow starts moving.
Run an example
ironflow run examples/00-showcase/nightly_report.lua